Glass Fall-Offs Detection for Glass Insulated Terminals via a Coarse-to-Fine Machine-Learning Framework.

Area of Science:

• Materials Science
• Electrical Engineering
• Computer Science

Background:

• Glass-insulated terminals (GITs) are crucial for high-reliability microelectronic systems.
• Glass fall-offs in GITs can compromise microelectronic component and device reliability.
• Automatic defect inspection is difficult due to light reflection, irregular defects, and limited data.

Purpose of the Study:

• To develop an automated inspection method for detecting glass fall-offs in GITs.
• To address the challenges of automatic defect detection in industrial settings.
• To improve the reliability assessment of microelectronic components.

Main Methods:

• A coarse-to-fine machine learning framework was proposed for glass fall-off detection.
• An adaptive sector partition scheme utilized the circular-ring geometry of GITs.
• Gradient Boosting Decision Trees (GBDT) and sector neighbor (SN) features were employed for classification.

Main Results:

• The proposed method achieved high performance on real industrial GIT images.
• Achieved an average Intersection over Union (IoU) of 96.85% and an F1-score of 0.984.
• Demonstrated a low pixel-level false alarm rate (0.55%) and a practical inspection speed (32.18 s/image).

Conclusions:

• The developed machine learning framework effectively detects glass fall-offs in GITs.
• The method offers a significant improvement over existing inspection approaches.
• This contributes to enhanced reliability and quality control in microelectronic manufacturing.